[ http://opensource.atlassian.com/projects/hibernate/browse/HSEARCH-1006?pa... ] Sanne Grinovero commented on HSEARCH-1006: ------------------------------------------ I don't really see how we can guarantee consistency between database and index by just taking an index snapshot, as some work might be "in the pipeline" and not applied yet. We can do this, but the only safe solution is to rebuild the index via the MassIndexer.. unless we allow to lock/freeze all transactions and flush the backends. -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira

[ http://opensource.atlassian.com/projects/hibernate/browse/HSEARCH-1006?pa... ] Sanne Grinovero commented on HSEARCH-1006: ------------------------------------------ Even flushToIndexes would just guarantee the current Session related work is sent to the backend, but there is no guarantee about the other sessions nor any guarantee at all for async backends. About locking.. I guess we could have all threads wait in a before-commit phase? If the goal is to make strong guarantees on the consistency of the index vs the database, this is definitely not trivial; if a slightly inconsistent version was acceptable one wouldn't need a snapshot and could just copy any index state. Sorry I don't like to sound pessimistic, but I don't see how to implement this.. unless you just mean to make it "slightly better" by freezing a snapshot: likely it's going to be a closer-in-time snapshot than just making a copy, but still people must be able to live with the fact it's not a perfect match. -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira